Party Institutionalization

The literature on parties distinguishes two key aspects of party systems – their internal and external characteristics. The external dimension is often referred to as ‘party system institutionalization’ (for example, Hicken, Kuhonta and Weiss Reference Hicken, Kuhonta and Weiss2015), highlighting features such as how stably aligned voters and legislators are to particular parties, for instance using measures of vote volatility. This, in turn, is partly a function of a country's institutional set-up, notably its electoral system, but also its broader regime type and level of economic development (see, for example, Mainwaring and Torcal Reference Mainwaring, Torcal, Katz and Crotty2006), and democratic history (for example, Mainwaring and Zoco Reference Mainwaring and Zoco2007). Further, extant research highlights how such linkages between parties, and especially electoral competition between them, can foster more inclusive social policies (for example, Fairfield and Garay, Reference Fairfield and Garay2017; Pribble Reference Pribble2013).

However, we focus on the internal dimension, often referred to as ‘party institutionalization’, which can be high even when electoral competition between parties is absent (as in contemporary China or Vietnam). In order to understand party institutionalization, it is helpful to start by considering the ‘positive pole’ of this multi-faceted concept – that is, a political party that is highly institutionalized in all relevant concerns. In this idealized situation, decisions within parties are taken according to clear, stable rules, and informed through dense networks linking party elites with broad constituencies outside the core organization. Hence, the following features contribute to a high degree of party institutionalization: (1) decision making, which may be more or less centralized to national-level party organizations or local entities, should follow clearly specified rules that allow for the transparent aggregation of preferences and arriving at unequivocal decisions, (2) well-specified and organizationally determined roles for decision makers, (3) hierarchical arrangements that permit disciplining actors (such as members of parliament (MPs) or local party elites) who stray from the party line, (4) organizational complexity that allows for the division of labor and effective outreach to different geographical areas, and (5) relatively stable linkages between the party and politically relevant mass constituencies, such as particular social groups, ethnic groups or citizens/voters located in certain geographic regions (see also Bizzarro et al. Reference Bizzarro2018).

Optimally, measures of party institutionalization should reflect this broad and multi-faceted concept, as well as the fact that parties could score relatively low/high on different features. More specifically, they should capture different features of party organizations, notably those listed directly above with points 1, 2 and 4, how parties function in co-ordinating and disciplining members (as indicated by point 3), and how parties link up to mass constituencies outside the core organization (as indicated by point 5), for example with enduring linkages expressed through stable, organizational co-operation with civil society organizations or widespread party membership. When parties lack these traits – that is, when party institutionalization is low – decision-making power rests outside the party organization, typically in the hands of particular individuals or narrow social groups. These individuals could be central party leaders. Hence, when decision-making power is centralized and located in a person rather than in the organization and its specified rules, party institutionalization is low.Footnote ² One example is ‘personalized rule’ (for example, Geddes Reference Geddes1999) where leaders can sometimes exercise power in a near-monopolistic manner despite formally belonging to a party (or absent any political party).

Before tying the concept of party institutionalization to welfare state features, we discuss, more generally, why certain political actors have stronger incentives and capabilities than others to erect encompassing welfare states that have programs with universal coverage.

Parties’ Incentives and Capabilities to Pursue Universal Social Policies

Citizens face various types of economic risks, particularly as workers: since most people are risk averse, they often want policy makers to alleviate these risks (Moene and Wallerstein Reference Moene and Wallerstein2001). Such demands can, however, be met by quite different policies. For example, risks related to involuntary unemployment can be mitigated by local constricted schemes, such as municipality-level workfare programs, as well as national schemes. Below we discuss how political parties vary in their incentives to create national welfare states that encompass multiple risks.

Yet even if parties decide that a nationwide program is desirable, the choice of who is to benefit remains. Should the program only cover workers in certain industries, be means tested and strictly targeted to the poor, or be ‘universal’ and cover all citizens?Footnote ³ When parties depend on narrow interests to stay in office, a relatively effective strategy is to introduce particularistic policy measures that channel resources to these groups instead of universal measures (which also advantage large groups of politically irrelevant citizens. See, for example, Bueno de Mesquita et al. Reference Bueno de Mesquita2003). Parties with broader constituencies should therefore prefer more universal programs than those that rely on narrower constituencies, be it geographically – in which case local programs may be preferred – or particular social groups – in which case national programs may be targeted to these groups (for example, Haggard and Kaufman Reference Haggard and Kaufman2008; Knutsen and Rasmussen Reference Knutsen and Rasmussen2018; Pribble Reference Pribble2013; Shefter Reference Shefter1977).Footnote ⁴

Adding to the relevance of different incentives to pursue particular welfare programs, political actors’ capacity to develop and implement effective policies matters. While a party might want to focus on national solutions to work–life risks, it could still be unable to pursue such policies effectively. This could stem from either a lack of capacity to control the behavior of key party members, who sometimes face strong incentives to stray from the party line (Ansolabehere, Snyder and Stewart Reference Ansolabehere, Snyder and Stewart2001), or an insufficient organizational apparatus for eliciting information about what viable policy designs look like in practice.

Given these hurdles, we cannot assume that all parties will favor, or have the ability to handle, various citizens' demands for risk-mitigating schemes through national-level universal social policies (see also, for example, Hicken Reference Hicken2009). We now detail our argument on how differences in party institutionalization contribute to explain variation in parties’ preferences and capacities to pursue encompassing and universal welfare policies.

Why Institutionalized Parties Lead to Welfare State Development

In the following, we explain how different facets of party institutionalization contribute to enhance parties’ incentives and capacities to pursue universal welfare policies pertaining to different areas of risk. While multiple relevant mechanisms tie party institutionalization to welfare policies, we simplify the presentation by grouping them into two categories. The first category, top-down mechanisms, addresses how institutionalized parties can create comprehensive bargains in order to arrive at credible policy proposals and override politicans or social groups aspiring to veto the implementation of inclusive (and expensive) welfare legislation (for example, Ehrlich Reference Ehrlich2007; Huber and Stephens Reference Huber and Stephens2001). The second category, bottom-up mechanisms, focuses on how institutionalized parties identify and aggregate demands from citizens and organized interests through civil society linkages or local branches.

The top-down mechanisms relate to how party institutionalization affects parties’ ability to discipline individual party members as well as overpower – or, perhaps even more important, strike credible deals with – powerful social groups that oppose the introduction of welfare schemes. Key in this regard is a strong party organization, which has the tools to monitor and sanction party members (inside and outside parliament) and other collaborators, as well as clear rules that permit the transparent aggregation of preferences from different actors in the party (for example, Janda and Colman Reference Janda and Colman1998; Tavits Reference Tavits2012a). In many political systems, several politicians or social groups have the standing to de facto veto legislation or the implementation of new policies that they dislike (for example, Ehrlich Reference Ehrlich2007; on insituitional veto, see Immergut Reference Immergut1992). For example, large-scale landowners stand to lose from the enactment of an unemployment benefits system, since they would expect to bear much of the associated tax burden but enjoy few of the benefits (Ansell and Samuels Reference Ansell and Samuels2014). Individual MPs with a formal party affiliation, but with an independent power base and diverging policy preferences, can also act as veto players. When facing such MPs, who may prefer targeted policies that benefit their personal constituency, the ability to call on party discipline is key for enabling national policy solutions.

More generally, how can parties, which may gain electoral or other benefits from pursuing welfare policies, overcome such veto players? We believe there are at least three mechanisms through which institutionalized parties can overcome resistant social groups or individual politicians with strong incentives to stray from the party line (see also Bizzarro et al. Reference Bizzarro2018). First, institutionalized parties should avoid renegade politicians by screening candidates according to, for example, ideological position and opportunistic tendencies. Well-established, party-centered selection processes, focusing on shared ideology, should help to ensure party unity (Carreras Reference Carreras2012; Janda and Colman Reference Janda and Colman1998). Second, institutionalized parties are enduring; insofar as this is known by relevant actors, actions should be taken with an eye towards their long-term consequences. When this second mechanism is combined with a third one – well-functioning organizational apparatuses provide fora for bargaining and tools for subsequent monitoring of different actors – party leaders, individual politicians and other powerful actors can strike comprehensive deals in which benefits and costs are distributed over longer time periods (Bizzarro et al. Reference Bizzarro2018; Boix and Svolik Reference Boix and Svolik2013; Gerring and Thacker Reference Gerring and Thacker2004; Hicken and Simmons Reference Hicken and Simmons2008).

Thus even if individual parliamentarians, for example from wealthy, rural districts, oppose comprehensive welfare legislation, a stable institutionalized party may allow party leaders to bring such politicians on board by credibly promising other policy or personal gains in mutually beneficial bargains. In a party with clear rules, and in which decision-making power is not centralized in a single person (national leader) or group of people (local party bosses), it is more difficult for actors to later renege on such bargains. This, in turn, makes the bargains more credible. Institutionalized parties should thus make politicians better able to overcome veto points and build broad coalitions behind universal welfare policies, covering multiple areas of risk, than less institutionalized parties.

In democracies (and perhaps even in some other electoral regimes with less intensive competition), institutionalized parties adopting long time horizons may also affect whether voters perceive future welfare payments as credible (Tavits Reference Tavits2012b, 84–85). If voters can choose between receiving an immediate good (reduced taxes, discretionary short-term payments, etc.) or an insurance against future risk (welfare benefit), the credibility of promises of the latter is key (Iversen Reference Iversen2005). If parties are unable to carry out policy promises, or are easily swayed to shift policy, voters will have few incentives to vote for parties that promise future welfare benefits. This suggests that institutionalized parties (in democracies) have stronger electoral incentives to expand welfare programs than non-institutionalized parties. In turn, voters are less likely to support universal welfare programs when parties are unable to credibly commit to promised programs, instead preferring personalized local goods such as jobs or direct cash payments, or some form of segmented arrangments instead (Kitschelt and Kselman Reference Kitschelt and Kselman2013; Pribble Reference Pribble2013, 31–32).

Regarding bottom-up mechanisms, parties differ in how well they predict and aggregate voter preferences (for example, Spies and Kaiser Reference Spies and Kaiser2014), in how they recruit candidates from different social strata across the territory (Katz Reference Katz2001; Jones and Mainwaring Reference Jones and Mainwaring2003; Tavits Reference Tavits2012b), and in how they form linkages with parts of the electorate or civil society organizations (Allern and Bale Reference Allern and Bale2012; Duverger Reference Duverger1959; Katz and Mair Reference Katz and Mair1995; Pribble Reference Pribble2013, 31). These differences influence the policies that parties pursue. We highlight how institutionalized parties establish local branches and stable and dense linkages with civil society organizations. Local party branches and strong linkages with society are vital for effectively catching and interpreting the demands of broad groups of citizens and for cultivating mass party identification (Samuels and Zucco Reference Samuels and Zucco2015; Tavits Reference Tavits2012b, 84–85). For example, local youth organizations can be set up to facilitate the training and selection of ideologically fitting candidates or candidates from under-represented groups (Folke and Rickne Reference Folke and Rickne2016).

While institutionalized parties in both autocratic and democratic contexts may establish local party branches and linkages, the specific motivations for doing so could differ. A core motivation of linking up with civil society under autocracy could, for example, be to co-opt potentially revolting industrial workers, whereas voter mobilization could be the motivation under democracy (Kim and Gandhi Reference Kim and Gandhi2010, 648).

We highlight that the motivation and more specific processes through which (and the partnering organizations with whom) such linkages form may also depend on party ideology. Considering the historical experiences of Western Europe, for example, social democratic parties formed close ideological affinities with trade unions. Unions helped parties with funding, mobilizing supporters, fostering a working class identity and shaping policy positions (Ebbinghaus Reference Ebbinghaus1995). Similarly, confessional parties, later Christian democratic parties, formed close linkages with various civil organizations (formed at the behest of the church), trade unions, employer associations and youth organizations (Kalvyas Reference Kalvyas1996, 63–108; Kalyvas and Kersbergen Reference Kalyvas and Kersbergen2010, 187). These linkages, which fostered greater institutionalization of some left-wing and Christian democratic parties of Western Europe, may, in turn, have pushed other parties in these countries to form extensive linkages with other organizations and voter groups in order to be able to compete with the former in terms of resources and votes (for example, Duverger Reference Duverger1959). Consequently, broad linkages can be built between civil organizations and parties of various ideological stripes, even if some types of parties have historically taken the lead in particular regions of the world (Duverger Reference Duverger1959; Kalvyas Reference Kalvyas1996).Footnote ⁵

Regardless of the motivation and more specific trajectory of linkage building, parties that construct such systems to elicit preferences may effectively register policy preferences and other relevant information from a broad array of groups and organizations (Duverger Reference Duverger1959; Kitschelt and Kselman Reference Kitschelt and Kselman2013; Samuels and Zucco Reference Samuels and Zucco2015). Institutionalized parties with local organizations should be better able to separate signals from noise, and their information-processing capacities make them less likely to focus too much on one particular signal (such as the expressed preferences of an important lobbying group or a particular bloc of voters in the capital). Local party meetings allow party representatives to hear the preferences of their constituents, discuss the party line and inform voters how the party understands various issues. Local branches also allow the party to recruit from a broader pool of citizens. Local branches will be staffed with organizers and candidates from the community, who, in turn, may rise in the central organization (Frendreis, Gibson and Vertz Reference Frendreis, Gibson and Vertz1990), and thus help the national organization form linkages with local civil organizations (Samuels and Zucco Reference Samuels and Zucco2015).

Further, when parties create strong national organizations and extensive local branches, they become identifiable focal points for various groups of citizens (Tavits Reference Tavits2012b, 85). This enables parties to interact with economically disadvantaged groups that are often less well organized, such as land workers, unskilled urban workers in service industries or women. These relatively poor groups, who are unlikely to bear the brunt of the tax burden of welfare programs, should prefer social policies encompassing multiple areas of risk and universal coverage. When parties have weak preference-aggregation systems, such as under cartel parties (Katz and Mair Reference Katz and Mair1995), cadre parties (Duverger Reference Duverger1959) or when parties are absent, the voices of such less resourceful citizens or those in distant regions might go unnoticed.

Highly institutionalized parties may thus try to represent the preferences of different local and social groups, and, at the same time, be capable of doing so with national policy solutions (Hicken, Kollman and Simmons Reference Hicken, Kollman and Simmons2016). The same informational capacity that allows parties to absorb information from different sources should allow institutionalized parties to transform these demands into policies that are amenable to effective implementation, for example by taking into consideration different practical obstacles to the registration of beneficiaries and monitoring of payments on the ground. Again, local constituency meetings, locally recruited candidates and links to civil society organizations, including unions, can inform the party of policy design issues or implementation problems. While high institutionalization gives parties the capacity to link up with broader groups of citizens, institutionalized parties may not always have an equally strong incentive to take the demands of these citizens into account. Given the need for broad support to win re-election in competitive multiparty contests, we thus anticipate that these bottom-up mechanisms will be particularly strong for institutionalized parties in democracies. We return to this more nuanced expectation below. Nonetheless, the wider discussion suggests two general hypotheses:

The welfare state literature primarily focuses on highly developed Organization for Economic Co-operation and Development (OECD) democracies (see Haggard and Kaufmann Reference Haggard and Kaufman2008). Much of this literature has focused on working-class or confessional parties as agents of welfare state development, and is inextricably linked to the historical rise of left and Christian democratic parties in these countries.Footnote ⁶ Yet the differences in party organization, stability and links to voters that we have highlighted differ, and should matter, within both developed and developing countries.

The features of parties that we focus on also differ within regime categories. But, as discussed, the proposed bottom-up mechanisms presumably operate more strongly in democracies, as the fortunes of politicians are more clearly linked to demands from broad groups of constituents through contested elections. Still, many autocratic regimes hold multiparty elections, and although these elections are not always contested (for example, Levitsky and Way Reference Levitsky and Way2010), autocratic regime parties still care about mass demands for various reasons, including the need to mitigate revolutionary threats (see, for example, Acemoglu and Robinson Reference Acemoglu and Robinson2006; Kim and Gandhi Reference Kim and Gandhi2010; Miller Reference Miller2015). Even if demands from large constituencies are less pertinent to some autocrats, especially those who do not face minimally competitive multiparty elections, all regimes rely on the support of some social groups to stay in power. And providing pension and other social policy programs that cover these groups is an effective distributional strategy for staying in power for autocrats (Knutsen and Rasmussen Reference Knutsen and Rasmussen2018). Further, we fail to see clear reasons why the top-down mechanisms should differ much between democratic and autocratic settings; leveraging an institutionalized party organization to negotiate with and overcome potential veto players such as local elites in rural areas may be an equally important tool for pursuing welfare state expansion for, say, Chinese Communist Party elites as for party elites in any democratic government.

In sum, we expect party institutionalization to matter for welfare state development in autocracies, but even more in democracies. Similarly, we anitcipate that institutionalized parties are important for overcoming the policy inertia associated with numerous veto players. We therefore expect institutionalization to be more important in settings with more veto players and in institutional contexts that enhance the latter's influence over policy making.

We test the implications of our argument by employing samples that pool historical information from all countries with available information. Yet we also assess how generalizable our argument is by testing for several relevant contextual factors that could moderate the proposed relationships between party institutionalization and welfare state features.

Party Institutionalization Index

We employ the V-Dem Party Institutionalization Index (PI) as our key independent variable. PI is presented in detail and validated in Bizzarro et al. (Reference Bizzarro, Hicken and Self2017). Briefly, PI records features of the main parties in a political system (easing comparisons between, for example, single- and multiparty systems); assessments are thus made at the country level. PI aims to capture ‘(1) the scope of party institutionalization in a country, (2) the proportion of parties that reach a threshold of minimal institutionalization, and (3) variations in the depth of this institutionalization – focusing on the links parties establish with voters and elites’ (Bizzarro et al. Reference Bizzarro, Hicken and Self2017, 2).

PI covers more than 170 countries (Coppedge et al. Reference Coppedge, Gerring, Lindberg, Skaaning, Teorell, Altman, Bernhard, Steven, Glynn, Hicken, Knutsen, Marquardt, McMann, Miri, Paxton, Pemstein, Staton, Tzelgov, Wang and Zimmerman2016a; Coppedge et al. Reference Coppedge, Gerring, Lindberg, Skaaning, Teorell, Altman, Bernhard, Steven, Glynn, Hicken, Knutsen, Marquardt, McMann, Miri, Paxton, Pemstein, Staton, Tzelgov, Wang and Zimmerman2016b), with time series back to 1900. V-Dem indicators are typically coded by five country experts, who originally respond to questions on ordinal five-point scales (see Appendix Table A1 for question wording). Subsequently, the V-Dem measurement model, a Bayesian item response theory model, recovers interval-level scores on the latent dimension that these ordinal categories map on to. It does so by leveraging the ordinal-category responses – and several other pieces of information – by multiple country experts. By leveraging various types of ‘bridging’ information, such as expert scores on anchoring vignettes and cross-country coding by some of the experts, the model allows us to account for differential item functioning and ensure cross-coder consistency, as well as cross-country and inter-temporal comparability (Pemstein et al. Reference Pemstein, Tzelgov and Wang2017).

PI draws on several indicators that tap into different features of parties highlighted as relevant by our theoretical argument (Bizzarro, Hicken and Self Reference Bizzarro, Hicken and Self2017, 7–9). Party organization (v2psorgs) considers how many parties have permanent organizations. Local branches (v2psprbrch) considers the number of parties with permanent local party branches. Distinct platforms (v2psplats) concerns how many parties among those with representation in the legislature have publicly available, and distinct, party platforms (manifestos). Legislative cohesiveness (v2pscohesv) assesses legislative cohesion for represented parties, capturing the extent to which political elites submit to the position of their parties when voting on important bills. Finally, constitutency linkages (v2psprlnks) considers the most common form of linkage between parties and their constituents, across all major parties; clientelistic linkages are assumed to signal low party institutionalization and programmatic policy linkages high institutionalization.Footnote ⁷ Among these indicators, local branches and constituency linkages are particularly attuned to capture the local embeddedness and societial links that were relevant for the bottom-up mechanisms, whereas party organization and legislative cohesiveness reflect the organizational tools and internal discipline and unity addressed by the top-down mechanisms. While we test these indicators individually, and briefly discuss the disaggregated results below, our main focus is on the composite PI measure, thus potentially capturing different mechanisms as set out in our comprehensive argument.

PI is aggregated by summing across standardized versions of these five indicators, and then normalizing the resulting measure on a 0–1 scale using its cumulative density function. An additive index allows for partial substitutability between indicators, meaning that a low value on one indicator can be compensated, but only partly, by high values on another (Goertz Reference Goertz2006). The aggregation thus reflects our argument in that different aspects of party institutionalization, such as a national organization with clear rules for decision making, or stable links with mass constituencies, may have some independent effects.Footnote ⁸ At the same time, partial substitutability suggests that low scores on some indicators can also (partially) be offset by high scores on others in terms of overall PI scores.Footnote ⁹

Figure 1 shows that (average) PI increased throughout the twentieth century, but at different speeds – and starting from different levels – in different regions. Western Europe and North America have historically displayed comparatively high levels of PI, and Africa and the Middle East comparatively low levels. East and South-East Asia experienced intense increases after WWII, whereas Latin America experienced its sharpest increase in the 1980s. Eastern Europe and Central Asia have experienced several periods of declining PI. The pattern is even more mixed for between- and within-country variation. For example, several countries have experienced (gradual or steep) declines over some periods of time and increases during others.

Figure 1. Average score on PI, over time, in eight world regions

Figure 2 displays PI scores for four countries with very dissimilar trajectories. Norway has had high and fairly stable scores across the time series. The Philippines, by contrast, has experienced relatively low PI scores, especially during American colonial rule and Japanese occupation, but also decades later during Marcos' strong-man rule. While PI spiked with democratization in 1986, it has remained comparatively low, and dropped very recently. Botswana had extremely low PI under British colonial rule, but experienced a dramatic increase with decolonization in 1966. The high score has persisted thereafter, under multiparty elections and (continued) Botswana Democratic Party government. Bulgaria experienced increasing, and very high, PI with communist rule after WWII. PI then declined with the introduction of multiparty politics.

Figure 2. PI score over time in four selected countries

Measures of Welfare State Features

We have argued that party institutionalization should foster national, extensive and universal social policies. There is no welfare state measure that properly captures all these dimensions together, except for over shorter periods of time (Scruggs Reference Scruggs2006). Instead, we employ several measures, from various sources.

Our first measure is from the SPaW dataset. It captures how encompassing welfare states are. Encompassingness counts whether there is a major, national welfare law for each of the following risks: old age, unemployment, maternity leave, child birth (family allowances), work injury and sickness. A major program implies that at least one of the following groups are covered: agricultural workers, industrial/production workers, small-firm workers, self-employed, students, employers, temporary/casual workers and family/domestic workers (see Knutsen and Rasmussen Reference Knutsen and Rasmussen2018). Encompassingness thus ranges from 0 (no major program) to 6 (major program in all areas). For the 9,053 observations in Model 1, Table 1, the mean score is 3.4 and the median is four programs. Figure 3 shows the distribution on Encompassingness, sorted by quartile on our measure of PI. Countries with high PI scores typically have more major welfare programs.

Figure 3. Histograms on Encompassingness for 9,053 obs. in Model 1, Table 1, by PI quartile

Table 1. Party institutionalization and welfare state encompassingness, universality, generosity and size: main models

Note: OLS with panel-corrected standard errors. T-values reported in parentheses. Constant, year dummies and country dummies not displayed. Concerning the dependent variable specifications, L = Level, C = Change (from t-1 to t). *p < 0.05, ** p < 0.01, ***p < 0.001

In order to capture how universal the welfare benefits are, we first use a universalism indicator from V-Dem: Universal Programs measures ‘[h]ow many welfare programs are means tested and how many benefit all (or virtually all) members of the polity?’ Beneficial for isolating the universalism dimension, V-Dem expert coders are explicitly told not to score whether there is a welfare state presence, but instead the structure of the existing benefits. Thus Universal Programs should not tap welfare state size per se; neither should it be related, by construction, to our measure of encompassingess. It is perfectly conceivable to have national-level social policy programs that cover multiple areas of risk but only channel benefits to, say, one social group.Footnote ¹⁰ One benefit of universal programs relative to other measures of welfare program features, and especially those strictly focused on de jure characteristics, is its substantial time-series variation, which captures the fact that several countries have experienced periods of substantial decline in the extent to which welfare benefits are universally provided (such as Bulgaria, Chile, New Zealand and Russia).

Overall, for the 16,267 observations in Model 4, Table 1, the mean, median and standard deviations of this universalism measure are, respectively, −0.17, 0.02 and 1.47. The lowest score is −3.26 and the highest is 3.23. Figure 4 displays histograms for these observations, sorting the sample by PI quartiles, and Figure 5 plots PI against Universal Programs for two selected years. In general, countries with higher PI have higher universalism scores.

Figure 4. Histograms on universal programs for 16,267 observations in Model 4, Table 1, by PI quartile

Figure 5. Scatter plots of PI and universal programs in 1950 (top) and 2000 (bottom)

Encompassingness and Universal Programs are our two primary measures. Both allow us to include more than 150 countries and time series longer than 100 years. Yet given concerns that PI and Universal Programs are both drawn from V-Dem and the potential resulting biases from combining two V-Dem measures that rely on expert coding, which we discuss more closely below, we test two additional measures of universalism.

The first is the Universalism Index (UI) from SPaW, which also has extensive coverage. UI counts the number of social groups, as listed above, for each of the six major risk/policy areas in SPaW (Rasmussen Reference Rasmussen2016). More specifically, 0 is given for a policy area if there is no major program; 1 if a program is means tested based on some property criteria (income-based exclusions are not considered means tested); 2 if one social group is covered by a contribution- or employment-based program; 3 if two groups are covered, etc. Finally, programs that automatically include all citizens are scored 9. Since there are six policy areas, and the programs are aggregated by addition, UI ranges from 0–54. The other measure, from the Social Citizenship Indicator Program (SCIP) (Korpi and Palme Reference Korpi and Palme2007), captures the share of the workforce that is insured against sickness in a state program. These data allow us to include only twenty-one countries (mostly ‘old’ OECD members), from 1930 to 2000. Appendix B report tests with this and other relevant measures.Footnote ¹¹

Control Variables and Benchmark Model

Our benchmark model is an ordinary least squares (OLS) estimator with panel-corrected standard errors. We always include country and year dummies. The country-fixed effects allow us to control for hard-to-observe factors that are fairly stable within countries (national party culture, ethnic composition, colonial history, a history of violent revolutions, etc.) and that may simultaneously determine both party institutionalization and social policy. Including country-fixed effects is made feasible by the long time series and substantial within-country variation in PI; the within-country standard deviation is 0.18, not far from the 0.26 between-country standard deviation. The year-fixed effects allow us to account for any global time trends in party and welfare state development and common ‘shocks’, such as International Labour Organization conventions, global economic depressions and world wars.

One complicating feature is the (presumably) complex causal relationships between PI and other observable factors that may affect welfare state development. Take income level, where one could expect that wealthier countries have advantages in allowing for more institutionalized parties (Kitschelt and Kselman Reference Kitschelt and Kselman2013) and better opportunities to finance welfare schemes. This suggests that we should control for income. Yet, extant work also finds that party institutionalization enhances economic development (Bizzarro et al. Reference Bizzarro2018), so including income could induce post-treatment bias by controlling for an indirect effect. Likewise, regime type and electoral system could also be post-treatment to political party features (see, for example, Kitschelt and Kselman Reference Kitschelt and Kselman2013).

Resolving this issue and identifying the single ‘true’ model is difficult. Our strategy is therefore to test both parsimonious models, which privilege mitigating post-treatment bias, and more extensive models, which privilege mitigating omitted variable bias. The most parsimonious models only include country and year dummies alongside PI as regressors. Our baseline extensive specification includes four controls for which we think there is a particularly strong theoretical rationale for considering as confounders. These are regime type/democracy, measured by V-Dem's Polyarchy (Teorell et al. Reference Teorell2019); electoral system (proportional representation and mixed-system dummies, with plural-majoritarian as the reference category) from V-Dem; ln GDP per capita (Purchasing Power Parity adjusted), from the Maddison Project (Bolt and Zanden Reference Bolt and Zanden2014); and ln population from Miller (Reference Miller2015). Our results, however, do not rely on using these two a priori preferred specifications. We report additional models, for example controlling for the lagged dependent variable, to further guard against unobserved confounders.

We also use military and public order spending as outcomes in placebo tests to mitigate concerns that PI simply correlates with all forms of state size. Further, we test models controlling for features such as urbanization, trade openness, income inequality, land inequality, union density, left-wing governments, interstate wars, civil wars, political corruption, civil society strength, female political participation and state–administration features, and we leverage generalized sensitivity analysis to assess how great omitted variable bias from other unobserved confounders must be for our results to be spurious.

Main Results

Encompassingness is our first dependent variable, capturing the number of national-level major welfare programs covering the six areas of risk as described above. Model 1, Table 1 is the parsimonious specification only including country- and year-fixed effects alongside PI. Drawing on 9,053 observations from 134 countries, it suggests a clear, positive significant relationship (t = 8.8). The point estimate suggests that a one-unit increase in PI increases the number of major national welfare programs by 0.59 – about one-sixth of the sample-mean number of programs (3.37). The result is only moderately attenuated when controlling for differences in electoral system, regime type, income and population in Model 2. While Model 2 suggests that high levels of per capita GDP, larger populations, proportional representation and – perhaps more surprisingly (but see Knutsen and Rasmussen Reference Knutsen and Rasmussen2018) – low levels of democracy also systematically relate to welfare state encompassingness, PI only drops from 0.59 to 0.42 and remains highly significant.Footnote ¹²

Despite the fixed effects and other controls, one might worry that other (time-variant) confounders or co-integrated trends generate a spurious link between PI and Encompassingness. Yet different specifications that should purge such potential confounding bias suggest that this is not so. Model 3, for instance, uses changes in Encompassingness from the previous year as the dependent variable and includes a lagged dependent variable as the regressor. Even in this conservative model, PI has a t-value of 2.6.

Models 4–6 replicate Models 1–3, but using Universal Programs as the dependent variable. As anticipated by our argument, PI is a clear and strong predictor.Footnote ¹³ Hence, the more institutionalized political parties are, the more that social policies are oriented towards broad groups. In Model 4, the estimated effect of going from the minimum (0) to maximum (1) values of PI is a 1.31-point increase in Universal Programs. This is only slightly lower than the sample standard deviation (1.47). The short-term relationship might even underestimate the effect. Indeed, when calculating the long-run effect from Model 6, we find an implied effect of about 2.55, close to a two-standard-deviation change in Universal Programs.

One issue – potential coder-induced bias – warrants caution when interpreting the results from Models 4–6. The results may be upward biased since some V-Dem coders are rating multiple V-Dem surveys and thus scoring countries on both of the indicators included in PI and Universal Programs. Yet such coder biases, if present, should also affect V-Dem's electoral democracy measure (Polyarchy). Controlling for Polyarchy should therefore help purge the PI–universalism relationship of this bias. Encouragingly, the results are very similar when including and excluding Polyarchy. Further, the PI–universalism result also persists for other universalism measures.

Model 7 displays an extensive specification that is similar to Model 5, but using the UI measure from SPaW. While UI has its own limitations, notably that it does not capture and weight the size of the social groups covered by the major welfare state programs, any relationship with PI should be unaffected by the type of coder bias discussed above. PI remains significant in Model 7 (t = 4.8), suggesting that institutionalized parties correspond to social policy programs covering more social groups, although the link with PI is not as robust to changing the model specification as for our other welfare measures (see, for example, Appendix Tables A6 and A17). Other results (for example, Appendix Tables A4 and A5) on quite different measures, including sickness benefit coverage data from OECD countries, reinforce the robust relationship between PI and welfare state universalism.

We now turn to disaggregating PI. Gauging results based on the individual indicators included in PI helps inform us about which of the discussed mechanisms operate more strongly, as the indicators relate to different, theoretically relevant sub-components of party institutionalization. For presentational reasons, we mainly discuss how the individual indicators of PI relate to welfare state universalism. Thus, the regressions presented in Table 2 employ Universal Programs as the dependent variable, but similar specifications on alternative dependent variables are in Appendix A5.

Table 2. Disaggregating party institutionalization and estimating effects on welfare state universalism

Note: OLS with panel-corrected standard errors. T-values reported in parentheses. *p < 0.05, **p < 0.01, ***p < 0.001

For Universal Programs, four of the five components included in PI have a strong and consistent relationship. Model 6, Table 2 shows that this also holds when including all five indicators simultaneously, suggesting that different mechanisms linking PI to welfare state development operate simultaneously, in accordance with our comprehensive argument (and the choice of additive aggregation for PI, allowing for partial substitutability between indicators). Due to the moderate to strong positive correlations between the indicators, we put more trust in Model 6 than Models 1–5 when discussing the individual relationships. Thus, despite the potential for multicolinearity increasing estimate uncertainty in Model 6, we find it prefereable to, say, Model 4, as Distinct Platforms in Model 4 might be biased by picking up an effect of either Local Branches (bivariate correlation of r = 0.82) or Constituency Linkages (r = 0.58) on welfare state universalism.

In Model 6, Party Linkages displays the strongest relationship, closely followed by Permanent Organizations, Local Branches and Legislative Cohesion. The presence of local branches and types of linkages were key to the bottom-up mechanisms posited in our theoretical argument, allowing institutionalized parties to widen their circle of constituents and elicit information from various groups. Nonetheless, the two features that are key to the top-down mechanisms are robust. Stable national organizations are vital for bargaining and overcoming various veto players, whereas legislative party cohesion is a direct signal of parties' ability to discipline and co-ordinate members.

Distinct Party Platforms, which presumably captures a clearly expressed ideology- and policy-based competition between parties, has the opposite effect of what we expected in Model 6, which simultaneously controls for all five indicators. This could be an indication that the correlation displayed in Model 4 is spurious, and that mere policy-based competition between parties is not a decisive factor for introducing universal benefits. This points to the finding, which we will discuss in the sub-section on assessing heterogeneity, that PI has a clear relationship with welfare state universalism in both democracies and autocracies.Footnote ¹⁴

Robustness Tests

We tested an extensive battery of specifications. Briefly summarized, the core finding (on PI) is very robust. Most tests are presented in the Appendix, but we report and discuss a selection of important, and quite different, tests in this section, focusing on welfare state universalism. Table 3 presents these tests for Universal Programs.

Table 3. Robustness tests on universal programs

Note: OLS with panel-corrected standard errors. T-values reported in parentheses. *p < 0.05, **p < 0.01, ***p < 0.001

Model 1, Table 3 replicates our extensive benchmark model (Model 6, Table 1). The first robustness test, Model 2, alters the lag specification, lagging all independent variables by five years. PI is somewhat attenuated, dropping from 1.3 to 0.8, but it remains highly significant. In Model 3, we further extend the lag to 10 years, and PI remains significant, although the estimated coefficient drops further to 0.3. These results suggest, first, that PI may have both fairly proximal and distal effects on welfare state development. Secondly, even though we cannot, of course, exclude the possibility, our core result is less likely to (only) reflect a reverse causality bias, as measuring the independent variable 10 years before the dependent variable should help mitigate X-Y circularity.

We also tested specifications with additional controls to further mitigate omitted variable bias (although this typically reduces the sample size and could introduce post-treatment bias). Models 4–6 are three examples. Model 4 includes extra socio-economic covariates that may affect universalism and correlate with PI – urbanization, trade openness, income inequality and land inequality (Appendix Table A1 provides operationalizations and data sources). Model 5 includes two pertinent controls for working-class organization – union density and left government. Model 6 adds controls for relevant national-level political institutions and events – interstate war, civil war, political corruption, impartiality of the state administration, civil society strength and participation in politics, and female political participation. These tests are important as they account for other core explanations of welfare state universalism – such as working-class power or female political mobilization (for example, Huber and Stephens Reference Huber and Stephens2001), or that having high-quality state institutions eases the implementation of comprehensive welfare programs (for example, Rothstein, Samanni and Teorell Reference Rothstein, Samanni and Teorell2012). Several of these controls are, indeed, highly significant in the expected direction. Yet our main result is stable. The benchmark PI coefficient is 1.3, whereas PI ranges between 1.3 and 1.6 in Models 4–6, and t-values range from 7.4 to 16.3. The results remain stable when we include different combinations of controls or enter any of them separately in the benchmark.

We tested various ways of aggregating the indicators (or subsets of indicators) of PI, such as extracting the first component from a factor analysis (Model 7). The result for welfare state universalism is robust. One specific worry is that the strong correlation could come from V-Dem coders considering whether parties adopt broad, national welfare programs and let this influence their coding of parties as programmatic rather than clientelist. Thus we add the linkages indicator as a separate control in Model 8, thereby purging the PI coefficient for the influence of this indicator. While PI is attenuated, both PI and linkages are sizeable and significant at 1 per cent. The result on welfare state universalism is also robust to dropping the linkages indicator before reconstructing PI (Appendix A8). We also note that the results are robust to using various cumulative measures of PI (for a similar measure on democracy, see Huber and Stephens Reference Huber and Stephens2012), thus capturing a country's recorded history of party institutionalization.

We also tested alternative estimators. Model 9, a random-effects model, allows us to also capture within-country variation. PI remains comparable in size to the benchmark, although the standard error increases. Nonetheless, PI remains significant at 1 per cent. In Model 10 we employ the ordinalized version of Universal Programs and re-estimate our benchmark using ordinal logit. PI remains robust.

We undertook two placebo tests, using military and public order spending as dependent variables (Appendix A9). We argued that PI is key to welfare state development because it incentivizes and enables parties to carry out national, universal legislative proposals even when particularistic elite interests oppose these initiatives. We surmise that military and public order spending do not face similar hostile vested (elite) interests, and we otherwise fail to see how our argument should pertain clearly to such spending. Consequently, we should not observe a clear relationship with PI. Instead, a significant relationship might indicate that PI correlates with all forms of state expansion, and that the relationship of interest is driven by some unmeasured confounder. PI is not systematically related to military spending (t-value 0.39) or public order spending (t-value −0.63).

To further assess the likelihood that omitted confounders are driving our results, we test how serious such bias would have to be for the t-value of the PI coefficient to drop below 1.96, using generalized sensitivity analysis. We provide a closer discussion and interpretation in the Appendix. In brief, any omitted factor would have to correlate much higher with our welfare measures and PI than any of the theoretically plausible controls included in our regressions, such as democracy (Polyarchy), GDP, population or trade openness. This further increases our confidence that omitted variable bias is unlikely to drive the observed correlation.

Assessing Heterogeneity

We now test whether our estimates mask heterogeneous effects across contexts by running split-sample tests for theoretically interesting factors. This is one simple and transparent way of investigating heterogeneity, but Appendix A10 and Figure 6 report interaction models as an alternative way of testing such heterogeneity. In general, we find systematic links between PI and the different welfare state features in various contexts. Again, we mainly focus on results for Universal Programs, and display other dependent variables in Appendix A7.

Figure 6. Estimated effect on universal programs of increasing PI from 0 to 1 in autocracies and democracies

Note: figure includes 95 per cent confidence intervals. The specification resembles Models 1 and 2, Table 4, but is run on full sample and includes interaction term.

Table 4. Split-sample tests on party institutionalization and universal programs

Note: OLS with panel-corrected standard errors. T-values reported in parentheses. Regime samples are determined by the Boix, Miller and Rosato (Reference Boix, Miller and Rosato2013) binary measure of democracy. For high- and low-inequality samples we use median sample-score (33) on percentage share of farm land that is family farms (from Vanhanen Reference Vanhanen1997) as cutoff. p < 0.05, **p < 0.01, ***p < 0.001

First, we noted that the proposed bottom-up mechanisms might not work as strongly in autocracies as in democracies, although the top-down mechanisms suggest that institutionalized parties should enhance, for example, universalism in both democracies and autocracies. Splitting the sample according to the dichotomous Boix, Miller and Rosato (Reference Boix, Miller and Rosato2013) measure (BMR), we find that PI is about three times larger in democracies (Model 2, Table 4) than in autocracies (Model 1). Nonetheless, PI is also substantially large and significant for autocracies. This difference across regimes is persistent, but typically smaller, in other specifications; Figure 6 presents point estimates from an interaction specification (resembling Models 1 and 2, but adding PI*BMR) run on the full sample. PI is a significant predictor in both regime contexts, but the estimated effect is about twice as large in democracies.Footnote ¹⁵

Secondly, we investigate whether the link between PI and universalism is stronger in contexts that offer ‘status quo actors’ greater leverage. PI should be more important for welfare state development in such contexts due to its importance for over-riding potential veto players. We use the Polcon V Index (Henisz Reference Henisz2002), which combines information on the number of independent government branches, control over these by a single party, judicial independence and sub-federal entities. Higher values equal greater institutional constraints on executives, presumably enhancing the leverage of veto players. Figure 7 shows the result from an extension of the benchmark, adding Polcon V and its interaction with PI. The results clearly suggest that PI has a larger marginal effect on welfare state universalism when veto players have greater leverage – while always positive and clearly different from zero, the estimated effect is much higher for high values on Polcon V.

Figure 7. Estimated marginal effect on universal programs of PI going from min to max on Polcon V index

Note: figure includes 95 per cent confidence intervals. The specification resembles the benchmark on full sample, but includes Polcon V and its interaction with PI.

Thirdly, we test whether our results are driven by the experiences of Western countries. If so, the support for our (general) argument would be weakened, or this would at least suggest some (untheorized) historical or cultural prerequisites for our mechanisms to work. While the PI coefficient is larger in Western (Model 3; 2.3) than other (Model 4; 1.0) countries, it is substantial in size and highly significant in both samples. (Appendix A10 contains analysis removing specific regions from the sample, and the relationship is robust). We also took out colonies and controlled for (10-year) post-independence effects, and the results are robust.

Fourthly, we checked whether the effect is strongly contingent on a society's socio-economic features. One might hypothesize that strong, mass-based parties help ensure the formulation and effective implementation of universal welfare programs only where there is low economic inequality, and thus smaller redistributive burdens on elites (see Boix Reference Boix2003). Alternatively, institutionalized parties could display a stronger relationship with universalism in high-inequality settings, as institutionalized parties would be needed to overcome powerful elite groups, such as rural landlords. Yet, when using Vanhanen's share of family farms as a proxy for land inequality, and splitting the sample at the median, PI is related to universalism in both low- and high-inequality settings; the coefficient is only moderately higher in low-inequality settings. The results are basically similar if we split the sample using Gini coefficients on income inequality (Table A34). Nor is the relationship restricted to developing or developed countries. When splitting the sample by median GDP per capita, PI relates quite similarly (and significantly) to Universal Programs in both contexts. Additional tests show fairly consistent relationships across countries that are less or more corrupt or that have less or more impartial public administrations, suggesting that strong state institutional capacity is not a prerequisite for institutionalized parties to influence welfare state development (cf. Shefter Reference Shefter1977).